Skip to content

Installing Nova

Nova is a prompt pattern matching framework designed to detect potentially harmful or problematic prompts for Large Language Models (LLMs). This guide explains how to install and configure Nova.

Installation

Nova is available as a Python package via pip. It works with Python 3.8 and above.

Quick Installation

pip install nova-hunting

This command will install Nova and its dependencies, including the novarun command-line tool which will be automatically added to your path.

Development Installation

For development or to get the latest version, you can install directly from GitHub:

pip install git+https://github.com/fr0gger/nova.git

Configuration

Nova requires API keys for the LLM providers you want to use. You can set these keys as environment variables:

Setting API Keys

OpenAI (Default)

# For OpenAI models (GPT-4, GPT-3.5, etc.)
export OPENAI_API_KEY="your_openai_api_key_here"

Anthropic

# For Anthropic models (Claude, etc.)
export ANTHROPIC_API_KEY="your_anthropic_api_key_here"

Azure OpenAI

# For Azure OpenAI Service
export AZURE_OPENAI_API_KEY="your_azure_api_key_here"
export AZURE_OPENAI_ENDPOINT="your_azure_endpoint_here"

Groq

# For Groq models (Llama-3, etc.)
export GROQ_API_KEY="your_groq_api_key_here"

Ollama (Local Models)

For Ollama, no API key is needed as it runs locally, but you need to have Ollama installed and running: 

# Optional: If Ollama is running on a different host or port
export OLLAMA_HOST="http://localhost:11434"

Configuration File

Nova also supports loading configuration from a YAML file. Create a file named nova_config.yaml:

# nova_config.yaml
llm:
  default: openai  # Default LLM provider
  openai:
    api_key: your_openai_api_key  # Overrides environment variable
    model: gpt-4o  # Default model to use
  anthropic:
    api_key: your_anthropic_api_key
    model: claude-3-sonnet-20240229
  azure:
    api_key: your_azure_api_key
    endpoint: your_azure_endpoint
    deployment_name: gpt-35-turbo
    api_version: 2023-05-15
  groq:
    api_key: your_groq_api_key
    model: llama-3.3-70b-versatile
  ollama:
    host: http://localhost:11434
    model: llama3

To use this configuration file with the Nova runner: 

novarun -r rules.nov -p "prompt" -c nova_config.yaml

Troubleshooting

Common Installation Issues

  1. Missing command-line tool: If the novarun command is not found, ensure that your Python binary directory is in your PATH. You can also run python -m nova.novarun as an alternative.
  2. Dependency conflicts: If you encounter dependency conflicts, consider using a virtual environment: 
    python -m venv nova-env
source nova-env/bin/activate  # On Windows: nova-env\\Scripts\\activate
pip install nova-hunting

LLM Connection Issues

  1. API key errors: Ensure your API keys are correctly set in your environment variables or configuration file.
  2. Ollama connection errors: If using Ollama, make sure the Ollama service is running and accessible.
  3. Network issues: Check your internet connection and firewall settings if you're having trouble connecting to external LLM providers.

Uninstallation

To remove Nova: 

pip uninstall nova-hunting

Next Steps

Once you have Nova installed, you can proceed to:

  • Creating Rules: Learn how to write detection rules
  • Running Nova: Start checking prompts against your rules